(342c) Combining Artificial Intelligence and Discrete Element Method Simulations for Oral Solid Dose Manufacturing Process Optimization - a Case Study on Bin Blending

The rapid and cost-effective design of reliable manufacturing processes is key to maximizing the profitability of pharmaceutical products, a large proportion of which are in the Oral Solid Dose (OSD) form. OSD manufacturing processes pose significant challenges because of the complex nature of the particulate materials they manipulate. The mechanics of these processes is still not fully understood leading to over-reliance on expensive and time-consuming physical trials in the process development stage where delays can be costly due to time limits on patents and competitive market conditions.

Virtual process design and optimization using Digital Twins (DTs) can reduce development timelines and costs and has become an important component of the digital transformation strategy in the pharmaceutical industry. Digital Twins based on high-fidelity Discrete Element Method (DEM) simulations have already been used to reduce manufacturing costs and time to market for new OSD products, but purely DEM simulation-based DTs can be computationally expensive and require expertise to develop and deploy. Going beyond the purely simulation driven approach to generate Artificial Intelligence (AI) based DTs from DEM simulation data can be a much more efficient approach.

In this work we present an efficient virtual optimization methodology for OSD manufacturing processes which combines DEM modelling, design of experiments, AI and optimization methods in the Altair portfolio of tools to significantly reduce the computational expense of optimization relative to a purely simulation driven approach. A key feature of the methodology is the generation of an AI based process Digital Twin, which is a multi-layer perceptron model that constitutes a state-space representation of a dynamic system and is trained on synthetic data from a statistically efficient set of high-fidelity DEM simulations. Because of its high computational efficiency, the AI based DT can be used in conjunction with a gradient descend optimization algorithm to identify the globally optimum operational parameter set in seconds rather than the weeks required by the equivalent purely simulation-based approach.

The validity of the methodology is evaluated through the optimization of an industrial scale bin blending system, which is commonly employed in OSD manufacturing operations. Batch failures are a persistent problem in this type of system, but its physical prototyping is expensive, making it an excellent candidate for virtual optimization. The work focuses on improving the mixing rate in the system by optimizing operational parameters such as the rotational velocity, the level of fill and the bin orientation. The accuracy of the predicted optimal operational parameter set is evaluated in silico, and the advantages and limitations of the methodology are discussed. Finally, ways to effectively deploy and democratize this technology through customized dashboarding are explored.